vnode_setup.in 7.23 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2
3
4

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2003 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
6
7
# All rights reserved.
#
8
9
10
11
12
13
14
15
16
17
18
19
20
use English;
use Getopt::Std;

#
# Set up the vnode state on a virtual (multiplexed) node.
#
# XXX - This script should only be run from os_setup!
#
# The output is all jumbled together since the updates are issued in parallel.
# Might be a pain when debugging. 
# 
sub usage()
{
21
    print STDOUT "Usage: vnode_setup [-f] [-k] <pid> <eid> [node ...]\n";
22
23
24
    exit(-1);
}
my  $optlist = "fdk";
25
26
27
28
29
30
31
32
33
34

#
# We don't want to run this script unless its the real version.
# That is, it must be setuid root. 
#
if ($EUID != 0) {
    die("*** $0:\n".
	"    Must be root! Maybe its a development version?\n");
}

35
36
37
38
39
40
41
#
# Configure variables
#
my $TB		= "@prefix@";
my $TESTMODE    = @TESTMODE@;
my $TBOPS       = "@TBOPSEMAIL@";
my $TBLOGS      = "@TBLOGSEMAIL@";
42
my $CLIENT_BIN  = "@CLIENT_BINDIR@";
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73

my $ssh		= "$TB/bin/sshtb -n";
my $debug       = 0;
my $force       = 0;
my $failed      = 0;
my $killmode    = 0;
my $mode        = "setup";
my $dbuid;

#
# Load the Testbed support stuff. 
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

# un-taint path
$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

# Turn off line buffering on output
$| = 1; 

#
# Parse command arguments. Once we return from getopts, all that should be
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
74
if (@ARGV < 2) {
75
76
77
78
79
80
81
82
83
84
85
86
    usage();
}
if (defined($options{"f"})) {
    $force = 1;
}
if (defined($options{"d"})) {
    $debug = 1;
}
if (defined($options{"k"})) {
    $killmode = 1;
    $mode = "teardown";
}
87
88
my $pid   = shift(@ARGV);
my $eid   = shift(@ARGV);
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108

#
# Untaint the arguments.
#
if ($pid =~ /^([-\@\w]+)$/) {
    $pid = $1;
}
else {
    die("*** Bad data in pid: $pid\n");
}	
if ($eid =~ /^([-\@\w]+)$/) {
    $eid = $1;
}
else {
    die("*** Bad data in eid: $eid\n");
}

#
# Verify permission to muck with this experiment.
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
109
110
111
112
if (!TBAdmin($UID) &&
    !TBExptAccessCheck($UID, $pid, $eid, TB_EXPT_DESTROY)) {
    die("*** $0:\n".
	"    You do not have permission to mess with $pid/$eid!\n");
113
114
115
}

#
116
# Get the list of nodes in this experiment.
117
#
118
119
my @nodes = ExpNodes($pid, $eid);
if (! @nodes) {
120
121
122
    warn("*** $0:\n".
	 "    No allocated nodes in experiment $pid/$eid!\n");
    exit(0);
123
}
124
125

# Nodes on the command line. Operate only on this set.
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
if (@ARGV) {
    my %fulllist = ();

    # Temporary hash list for searching.
    foreach my $node ( @nodes ) {
	$fulllist{$node} = 1;
    }
    @nodes = ();

    foreach my $node ( @ARGV ) {
	if ($node =~ /^([-\@\w]+)$/) {
	    $node = $1;

	    if (!defined($fulllist{$node})) {
		die("*** $0:\n".
		     "   Node $node is not allocated to $pid/$eid!\n");
	    }
	}
	else {
	    die("Bad node name: $node.");
	}
	push(@nodes, $node);
    }
}
my $exptstate = ExpState($pid, $eid);
151
152
153
154

# Just the vnodes mam.
foreach my $node (@nodes) {
    my $pnode;
155
    my $jailed;
156
    my $plab;
157
    my $allocstate;
158

159
    if (! TBIsNodeVirtual($node, \$jailed, \$plab)) {
160
161
	next;
    }
162
163
164
165
    if (! TBPhysNodeID($node, \$pnode)) {
	die("*** $0:\n".
	    "    No physical node for $node!\n");
    }
166

167
168
    TBGetNodeAllocState($node, \$allocstate);

169
170
171
    #
    # On remote nodes, or when forcemode is on, always do the deed.
    # Otherwise, look at experiment state.
172
173
    #
    if (!$force) {
174
175
176
177
	if ($exptstate eq EXPTSTATE_SWAPPING) {
	    #
	    # When swapping, local vnodes go down with the physnode.
	    #
178
179
180
181
182
183
184
185
	    if (! TBIsNodeRemote($node)) {
		print "$node will $mode with local node $pnode.\n";
		next;
	    }
	    elsif ($allocstate eq TBDB_ALLOCSTATE_DOWN) {
		print "$node never booted; skipping.\n";
		next;
	    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
186
	}
187
188
	elsif ($exptstate eq EXPTSTATE_ACTIVATING) {
	    #
189
190
191
192
193
194
195
196
197
198
199
	    # The allocstate determines if the vnode actually needs to
	    # be setup or torndown. Note that a failed experiment will
	    # cause a bunch of vnodes to be torndown, while in the
	    # ACTIVATING state. See os_setup and assign_wrapper; the
	    # idea is to avoid doing setup/teardown up vnodes on
	    # machines that are rebooting anyway, or that failed.
	    # Complicated by modify which could add/subtract a vnode on
	    # an existing machine, but not reboot the machine. Note that
	    # free (now unused) vnodes will land in RES_TEARDOWN. It is
	    # assumed that these booted okay, and need to be torndown,
	    # even though they are not RES_READY.
200
	    #
201
202
203
	    if (! TBIsNodeRemote($node)) {
		if ($killmode) {
		    if ($allocstate eq TBDB_ALLOCSTATE_DOWN) {
204
205
206
207
			print "$node failed to boot; skipping $mode.\n";
			next;
		    }
		    elsif ($allocstate eq TBDB_ALLOCSTATE_RES_INIT_CLEAN()) {
208
209
210
			print "$node never booted; skipping $mode.\n";
			next;
		    }
211
212
213
214
		    elsif ($allocstate eq TBDB_ALLOCSTATE_RES_READY()) {
			print "$node will $mode with local node $pnode.\n";
			next;
		    }
215
216
		}
		elsif ($allocstate eq TBDB_ALLOCSTATE_RES_READY()) {
217
		    print "$node is already setting up on local node $pnode\n";
218
219
220
221
		    next;
		}
	    }
	    elsif ($killmode && $allocstate eq TBDB_ALLOCSTATE_DOWN) {
222
		print "$node failed to boot; skipping $mode.\n";
223
224
225
		next;
	    }
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
226
227
228
229
230
231
232
233
234
    }

    #
    # When setting up a vnode, force its event state into SHUTDOWN since
    # no telling what its initial state is. 
    # 
    if (!$killmode) {
	TBSetNodeEventState($node, TBDB_NODESTATE_SHUTDOWN);
    }
235
236
237
238
239
240
241
242
243
244
245
246
    
    print STDOUT "Doing $mode of vnode $node on $pnode ...\n";

    #
    # Run an ssh command in a child process, protected by an alarm to
    # ensure that the ssh is not hung up forever if the machine is in
    # some funky state.
    # 
    my $syspid = fork();

    if ($syspid) {
	local $SIG{ALRM} = sub { kill("TERM", $syspid); };
247
	alarm 60;
248
249
250
251
	waitpid($syspid, 0);
	alarm 0;

	#
252
	# Look for setup failure, reported back through ssh.
253
	# 
254
255
256
	if ($?) {
	    my $exitstatus = $?;

257
258
	    print STDERR "vnode $mode on $pnode returned $?.\n" if $debug;

259
	    if ($exitstatus == 256) {
260
		print STDERR "$node is not running sshd.\n" if $debug;
261
262
	    }
	    elsif ($exitstatus == 15) {
263
264
265
266
267
268
		print STDERR "$node is wedged.\n" if $debug;
	    }
	    
	    # Send mail to testbed-ops about it
	    SENDMAIL($TBOPS, "Virtual Node $node $mode failure",
	     "Virtual node $node $mode (on physical node $pnode) in pid/eid\n".
269
	     "$pid/$eid has failed! Exit status was $exitstatus.\n");
270
271
272
273
274
275
276
277
278
279
280
281

	    if (!$killmode) {
		die("*** $0:\n".
		    "    Virtual node $node setup failure!\n");
	    }
	    else {
		warn("*** $0:\n".
		    "    Virtual node $node teardown failure!\n");
	    }
	}
    }
    else {
282
283
	my $args = ($killmode ? "-k " : " ");
	$args   .= ($jailed   ? "-j " : " ");
284
	$args   .= ($plab     ? "-p " : " ");
285
	$args   .= "$node ";
286
287
288
	
	# Must change our real UID to root so that ssh will work.
	$UID = 0;
289
290
291
292
293
294
295
296
297

	if ($plab && !$killmode) {
	    if (system("$TB/sbin/plabnode -v -d alloc $pid $eid $node")) {
		die("*** $0:\n".
		    "    Plab node allocation failed");
	    }
	}

	exec("$ssh -host $node $CLIENT_BIN/vnodesetup $args");
298
299
	die("*** $0:\n".
	    "    exec failed!\n");
300
301
302
303
304
305
306

	if ($plab && $killmode) {
	    if (system("$TB/sbin/plabnode -v -d free $pid $eid $node")) {
		warn("*** $0:\n".
		     "    Plab node free failed");
	    }
	}
307
308
309
    }
}

310
311
312
313
314
315
if ($killmode) {
    print STDOUT "Vnode teardown finished.\n";
}
else {
    print STDOUT "Vnode setup initiated on all nodes ...\n";
}
316
exit 0;